Gemma 4: Google Open Text Encoder for ComfyUI

Gemma 4 is Google DeepMind's open text encoder model for ComfyUI, available in E2B and E4B instruction-tuned variants with BF16 and FP8 precision.

G

Gemma 4

Text EncoderMultimodalOpen SourceApache 2.0

Open model family by Google DeepMind. Gemma 4 handles text and image input with configurable thinking modes, supporting up to 256K context and over 140 languages. Packaged as ComfyUI text encoder checkpoints in instruction-tuned E2B (2.3B effective) and E4B (4.5B effective) variants with BF16 and FP8 quantization.

Hugging FaceGitHubBlogTech ReportDocumentation
DeveloperGoogle DeepMind
Release Date2025-07
ArchitectureDense Transformer with Sliding Window + Global Attention
LicenseApache-2.0
VariantsE2B IT (2.3B eff), E4B IT (4.5B eff)
Context Window128K tokens
Multilingual140+ languages
QuantizationsBF16, FP8 scaled (E4B only)
CapabilitiesText, Image, Audio (E2B/E4B), code, reasoning, function calling

Guides and workflows related to this model series.

No articles found.